Application Of Adaptive Noise Cancellation With Neural-Network-Based Fuzzy Inference System For Visual Evoked Potentials Estimation

Visual Evoked Potential (VEP) is a gross electrical response of the braintime-locked to an external visual stimulus. The Visual Evoked Potential (VEP) has avery important role in the field of neurological physiology and clinical diagnosis. Likemany biological signals, the recorded VEP is corrupted by noise from backgroundongoing activities~ A main source of noise is the spontaneous electroencephalogram(EEG) generated in the brain. Since the amplitude of this noise is much higher thanthat of the VEP response, the signal noise ratio (SNR) of recorded VEP is typicallylow. Ensemble averaging (EA) is a conventional method for improving the SNR ofthe recorded VEP. Depending on the SNR of the recorded signal, EA requires100-2000 trials to estimate an EP. Targets of others methods are concentrated ondecreasing the number of repetitions required and speeding up EP estimation. Theideal goal is to estimate EP in a single trial. WIDROW and G1OVER first approached adaptive noise cancellation (ANC) .The method uses a "primary" input containing the corrupted signal and a "reference"input containing noise correlated in some unknown way with the primary noise. Thereference input is adaptively filtered and subtracted from the primary input to obtainthe signal estimated. This paper presents an application of adaptive noise cancellationwith adaptive-network-based fuzzy inference system (ANFIS) for rapid estimation ofvisual evoked potentials (VEPs). Usually a recorded VEP is severely contaminated bybackground ongoing activities of the spontaneous EEG signal in human brain. Manyapproaches have been adopted to enhance the signal-to-noise ratio (SNR) of therecorded signal. However, nonlinear dynamic methods are rarely investigated in viewof their complexity, and the fact that the nonlinear characteristics of the signal arehard to determine in general~ An adaptive noise cancellation method with ANFIS wascarefully designed to estimate the VEP signal. ANFIS based on Takagi and Sugeno'sfuzzy model has the advantage of being linear-in-parameter; thus the conventionaladaptive methods can be efficiently utilized to estimate its parameters. Anotheradvantage of ANFIS lies in that it can track the dynamic behavior of VEP in areal-time fashion because the VEP variation tracking is important for critical patientmonitoring in the clinical situation. A series of computer experiments conducted onsimulated and real-test responses have confirmed the superiority of the methoddeveloped in this paper.